Chemical Engineering and Processing 52 (2012) 112–124

Contents lists available at SciVerse ScienceDirect

Chemical Engineering and Processing:

Process Intensification
journal homepage: www.elsevier.com/locate/cep

Economically efficient operation of CO2 capturing process. Part II. Design of

control layer
Mehdi Panahi, Sigurd Skogestad ∗
Department of Chemical Engineering, Norwegian University of Science and Technology (NTNU), 7491 Trondheim, Norway

a r t i c l e i n f o a b s t r a c t

Article history: In part I of this study, control structures were proposed for different operational regions of a post-
Received 24 February 2011 combustion CO2 capturing process using the top-down steady-state economic part of the plantwide
Received in revised form 18 July 2011 procedure. In the current study, the bottom-up part of the complete procedure is considered. For this
Accepted 8 November 2011
purpose, dynamic simulation is used to validate the proposed control structures. Different control config-
Available online 17 November 2011
urations using decentralized controllers and model predictive control (MPC) are considered. At the end, a
simple control configuration is proposed which keeps the process close to the optimum in all operational
Keywords:
regions without the need for switching the control loops.
Plantwide control
Stability © 2011 Elsevier B.V. All rights reserved.
Decentralized controllers
MPC
Large disturbances

1. Introduction as possible) that implements in practice the steady-state control

objectives from part I over the entire feasible throughput range. A
We study optimal operation of a post-combustion CO2 captur- main issue is to handle transition from region I to region II where an
ing process, where the objective is to minimize the sum of the unconstrained degree of freedom is lost due to a constraint becom-
energy cost and penalty cost for releasing CO2 to the atmosphere. ing active. In general, some logical switching or reconfiguration of
In part I of this work [1], a top-down analysis of the complete control loops would be necessary to manage the transition. Here,
plantwide control procedure (Table 1) was performed to identify using prior knowledge of the constraint that becomes active, we
different operational regions of active constraints as a function of synthesize a simple control structure that does not require any loop
the throughput (flow rate of flue gas) and to select self-optimizing reconfiguration and thus provides near-optimal operation over the
controlled variables (CVs) in each region. In region I, the flue gas entire feasible throughput range. The details on how we arrived at
flow rate is given at its nominal value and there are two uncon- the chosen pairings are the main issue for the present paper.
strained degrees of freedom (DOF), which may be considered as Towards the synthesis of such a simple structure, four different
the CO2 recovery in the absorber and the CO2 mole fraction at the control configurations (including the one in Fig. 1) using decentral-
bottom of the stripper. The best associated CVs were found to be ized PI controllers, briefly summarized below, are considered.
the CO2 recovery in the absorber and the temperature on tray no.16
in the stripper [1] (see Fig. 1). However, these CVs are not neces-
• Alternative 1: The two unconstrained self-optimizing CVs for
sarily the best in all operating regions and for larger flowrates of
region I are controlled using the most obvious pairings (Fig. 1).
the flue gas (region II), where the reboiler duty reaches its maxi-
• Alternative 2: The two self-optimizing CVs for region I are con-
mum, the temperature of tray 13 in the stripper was found to be
trolled using the reverse pairings compared to Alternative 1.
the best unconstrained CV [1]. For even larger flue gas flowrates,
• Alternative 3: The best self-optimizing CV for region II is con-
one reaches the minimum allowable CO2 recovery of 80% and we
trolled.
have reached the bottleneck where a further throughput increase
• Alternative 4: Recommended modification of Alternative 2 which
is infeasible (region III) [1].
provides near optimal operation over the entire throughput range
In this work, we focus on the bottom-up part of the procedure
(regions I and II).
in Table 1, where we want to identify a control structure (as simple

Also, the possibility of using multivariable control is considered.

∗ Corresponding author. Tel.: +47 73594154; fax: +47 73594080. To validate the proposed control structures, dynamic process
E-mail address: skoge@ntnu.no (S. Skogestad). simulation is performed in UniSim [3].

0255-2701/$ – see front matter © 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2011.11.004
 M. Panahi, S. Skogestad / Chemical Engineering and Processing 52 (2012) 112–124 113

Table 1 the capacity bottleneck it will reach the minimum allowed recov-
Plantwide control procedure [2].
ery which is set to 80%. The details about the objective function,
I. Top-down part (focus on steady-state economics) optimization and selection of the best self-optimizing controlled
Step 1. Define operational objectives (economic cost J to be minimized) and variables are given in part I of this study [1].
constraints
The article is organized as follows. In Section 2, design of the
Step 2. Identify degrees of freedom (MVs) and optimize the operation for
important disturbances (offline analysis) to identify regions of active
control loops for primary (CV1) and secondary (CV2) controlled
constraints variables in regulatory and supervisory control layers is presented.
Step 3. Each region of active constraints: select primary (economic) controlled In Section 3, alternative control structures to handle larger through-
variables CV1: puts are introduced and discussed. The dynamic performance of the
-Active constrains
alternative control structures is evaluated for large disturbances in
-“Self-optimizing” CV1s for the remaining unconstrained degrees of freedom
Step 4. Select location of throughput manipulator (TPM) Section 4 and the best control structure for (near) optimal operation
II. Bottom-up part (focus on dynamics) over the entire throughput range is recommended.
Step 5. Choose structure of regulatory control layer (including inventory
control)
a. Select “stabilizing” controlled variables CV2 2. Design of the control layers
b. Select inputs (valves) and “pairings” for controlling CV2
Step 6. Select structure of supervisory control layer
Step 7. Select structure of (or need for) optimization layer (RTO)
In general, the control system can be divided in two main layers
(see Fig. 2).
There are some other works that study the dynamic behavior of
the CO2 capturing processes ([4–7]) but the current study is the first • Regulatory layer. Control of secondary (stabilizing) controlled
to consider different operational regions and synthesizes a simple variables (CV2). This layer usually involves the use of single loop
control structure that works in the different regions. PID controller.
In another study [8], we designed a control structure for a CO2 • Supervisory (economic) control layer. Control of the primary (eco-
capturing process where the objective function was to minimize nomic) controlled variables (CV1) using as manipulated variables
energy requirement with fixed CO2 recovery (90%). In the current (MVs) the setpoints to the regulatory layer or “unused” valves
study, an economic penalty on the CO2 released to the air is further (from the original MVs). This layer is usually about a factor 10
imposed, which makes it optimal to remove higher amounts of CO2 or more slower than the regulatory layer. Since interactions are
(∼95%). However, at higher flue gas rates, when the capacity con- more important at longer time scales, multivariable control may
straint for the stripper is reached, the CO2 recovery will drop, and at be considered in this layer.

Fig. 1. Alternative 1, proposed decentralized control structure for region I with given flue gas flowrate [1].
114 M. Panahi, S. Skogestad / Chemical Engineering and Processing 52 (2012) 112–124

The details of the regulatory layer design (step 5) are given

in Table 2. In addition to the comments given in Table 2, one
would generally like to combine control tasks in order to simplify
the control system. For example, one may select to control the
same stripper tray temperature in both regions I and II although
there may be a small economic penalty. Furthermore, stabilizing
the stripper temperature profile is necessary to maintain the CO2
inventory circulating around the amine recycle loop. The stripper
temperature control loop may therefore be moved to the regulatory
layer.
Let us go back to the decisions involved in designing the control
layer (Table 2).
(a) The first issue is to identify the variables that need to be
controlled to “stabilize” the operation. Here “stabilization” means
that the process does not “drift” too far away from the designed
operational point when there are disturbances. For our process,
we identify the following seven “stabilizing” CVs which need to
be controlled (CV2):

1. Absorber bottom level,

2. Stripper (distillation column) temperature,
3. Stripper bottom level,
4. Stripper top level,
5. Stripper pressure,
6 and 7. Recycle surge tank: inventories of water and amine.

Note that there is no need to control the absorber temperature

for the purpose of stabilization its profile. However, the tempera-
ture inside the absorber needs to be kept at a given value for good
operation. We select to do this by controlling the absorber liquid
feed temperature at 51 ◦ C [1] in the regulatory layer.
The absorber pressure is not controlled (“floating”) because a
fixed value would require a valve which would give an undesired
loss. It is set indirectly by the ambient pressure.
In addition, the inventories of water and amine in the recycle
system must be controlled to make up for losses. However, these
are small so even manual control may be possible.
Fig. 2. Typical control hierarchy in a chemical plant [2]. (b) The next decision is to select the inputs “pairings” to control
these variables (CV2).
Let us consider operation in region I (Fig. 1). We start with the
Let us start with the control objectives for the supervisory
inventory control system, that is, control of levels and pressure.
control layer, which generally may change depending on the distur-
The feed flow is given so the inventory control system needs to be
bances and active constraints. As a result of step 3 in the procedure
in the direction of flow ([2] and [9]). However, for this particular
(Table 1), in region I (low feed rates) the primary (economic) con-
process this does not really fix any loops, except for the pressure
trolled variables were identified as [1]:
control of stripper using the CO2 outflow (V-3). Next, we consider
the closed recycle system of amine and water where we, as men-
CV11 = CO2 recovery in the absorber,
tioned, need to decide on where to set the recycle flowrate. We
CV12 = Temperature at tray 16 in the stripper.
choose to set it using the recycle liquid flow to the absorber col-
umn (V-8). The choice of V-8 as the flow manipulator in the recycle
In region II with higher feedrates, the stripper reboiler duty is at loop is an important decision (we will reconsider it later), because
maximum, so there is one less degree of freedom and the best CV the “radiation rule” (Table 2) implies that inventory control in the
was identified as [1]: recycle loop “downstream” of this location must be in the direction
of flow. Thus, the bottom levels in the absorber and stripper must
CV1 = Temperature at tray 13 in the stripper. be controlled by their outflows (V-1 and V-6 respectively).
The pump V-10 in the recycle controls the pump outlet pressure
Furthermore, before starting the bottom-up part of the proce- but this is mainly for simulation purposes and the pump could also
dure, which starts with the choice of the regulatory layer (step 5), be set to run on constant power. However, when the pump controls
one generally needs to locate the throughput manipulator (step 4). pressure, the pressure measurement must be after the pump (in
However, the CO2 capture plant is a part of the overall power plant, the opposite direction of flow). Finally, the stripper has a partial
and the throughput manipulator is located further upstream in the condenser, and the condenser level is controlled by the reflux flow
power plant. This means that the feed (flue gas) to the CO2 capture (V-2).
plant is given and will act as a disturbance, and the control sys- Next, with the inventory control system fixed, we consider the
tem must be set up to handle this disturbance. However, the CO2 stabilizing temperature loop for the stripper. We choose to use the
capture contains a closed amine/water system and one must set the reboiler duty (steam V-5) as the MV which is the only remaining
amine flow between the columns. The location of the amine recycle option. To combine regulatory and supervisory control, the tem-
flow manipulator is an important decision. perature sensor is located at tray 16, which was identified as a
 M. Panahi, S. Skogestad / Chemical Engineering and Processing 52 (2012) 112–124 115

Table 2
Details on step 5. Structure of regulatory (stabilizing) layer [2].

The purpose of the regulatory layer is to “stabilize” the plant, preferably using a simple control structure with single-loop PID controllers. “Stabilize” here means that
the process does not “drift” away from acceptable operating conditions when there are disturbances. In addition, the regulatory layer should follow the setpoints given
by the “supervisory layer”.
Reassignments (logic) in the regulatory layer should be avoided: preferably, the regulatory layer should be independent of the economic control objectives (regions of
steady-state active constraints), which may change depending on disturbances, prices and market conditions.

The main decisions are:

(a) Identify CV2s for the regulatory layer, these include “stabilizing” CVs, which are typically levels, pressures, reactor temperature and temperature profile in distillation
column. In addition, active constraints (CV1) that require tight control (small back-off) may be assigned to the regulatory layer.
(b) Identify pairings (MVs to be used to control CV2), taking into account:
–Want “local consistency” for the inventory control [9]. This implies that the inventory control system is radiating around a given flow.
–Avoid selecting as MVs in the regulatory layer, variables that may optimally saturate (steady-state), because this would require either
◦ Reassignment of regulatory loop (complication penalty), or
◦ Back-off for the MV variable (economic penalty)
– Want tight control of important active constraints (to avoid back-off).
– The general pairing rule is to “pair close” to achieve a small effective delay from input (MV) to outputs (CV2).

self-optimizing variable in region I. Finally, the temperature of the Pairing issues

absorber liquid feed is kept at 51 ◦ C using the cooler duty (V-9).
Let us finally consider the supervisory layer which operates at
Alternative 1 may seem to be an “obvious” pairing choice but a
a slower time scale. The stripper condenser temperature, which
more careful analysis shows that this is not so clear. A useful tool
should be low to reduce the required work for CO2 compression
for selecting pairings is the relative gain array (RGA) and two main
in the downstream process, should clearly be controlled using the
pairing rules are [10]:
cooling water flow (V-4). The remaining “economic” variable to be
RGA-rule 1. Prefer pairings such that the rearranged system, with
controlled is the CO2 recovery and as a MV we may use the recycle
the selected pairings along the diagonal, has an RGA matrix close
amine flowrate to the absorber column (V-8), which was not used
to identity at frequencies around the closed-loop bandwidth. This
in the regulatory layer. This will be a relatively slow loop. The final
may be quantified by selecting pairings with a small RGA number:
control structure for region I is shown in Fig. 1.  
RGA number = RGA(G) − I  (1)
sum
3. Alternative control structures to handle larger
RGA-rule 2. Avoid pairing on negative steady-state RGA ele-
throughputs
ments.
The second rule states that pairing on a negative RGA provides
In region II, the stripper reboiler duty (steam) is at its maximum
a potential unstable response when individual loops are malfunc-
so the stabilizing temperature loop used for region I (Fig. 1) will
tioning, as it is the case of input saturation. To compute the RGA we
no longer work. This is to be expected, because we have not fol-
need a dynamic model. Using the dynamic UniSim simulator and
lowed the recommendation in Table 2 that says “Avoid selecting as
“Profit Design Studio” (PDS) [11], we identified the following linear
MVs in the regulatory layer, variables that may optimally saturate”.
model:
The problem is that there are no obvious alternative for controlling ⎡ 6.85s + 1.74 −0.76s + 0.038
⎤
the column temperature. The reflux is often used for distillation
columns, but this is a stripper where the reflux of the key com- Gdyn. (s) = ⎣ 19.7s2 + 11.4s + 1 2400s2 + 107s + 1 ⎦ (2)
(−9.51s − 1.02)e−2s 0.45s + 0.0754
ponent (CO2 ) is very small and the reflux has no effect on column
temperature. In addition, from the steady state optimization, it is 218s2 + 17.3s + 1 205s2 + 18.8s + 1
optimal with maximum cooling to minimize the CO2 compressor The steady-state RGA computed from this model is
work in the down stream process [1]. Therefore, reflux must be  
0.77 0.23
used to control condenser level. Alternatively, the column federate RGAdyn. (s = 0) = (3)
0.23 0.77
could be used, but this is already used for controlling the absorber
sump level. Since all elements are positive one cannot eliminate any of the
pairings using RGA-rule 2. The RGA-number as a function of fre-
3.1. Alternative 2 (“reverse pairing”) quencies is plotted for the two alternative pairings in Fig. 4. As
expected, we find that the diagonal pairing is the best with the
To find alternative solutions, let us start with region I. We con- RGA-number close to 0 at all frequencies.
sider the problem of controlling the two self-optimizing CVs. To check the model obtained from the dynamic simulator, we
compute the steady-state gains using the steady-state UniSim
y1 : CO2 recovery model and this gave a very surprising result. By making 5% per-
y2 : temperature on tray no.16 in the stripper turbations in the inputs, the following steady-state model was
identified:
 
using the two available MVs −0.5232 1.48
GSS = 10−2 × (4)
−8.47 5.17
u1 : recycle amine flowrate (V-8)
u2 : reboiler duty (V-5). with the corresponding RGA matrix:
 
−0.27 +1.27
The pairing in Fig. 1, where y1 is controlled using u1 and y2 is RGASS = (5)
+1.27 −0.27
controlled using u2 is referred to as the “diagonal” pairing (Alter-
native 1). We will also consider the reverse “off-diagonal” pairing Note that steady-state RGA for the diagonal pairing is negative.
(Alternative 2; see Fig. 3). The reason is that the sign of the 1,1-element of the gain matrix
116 M. Panahi, S. Skogestad / Chemical Engineering and Processing 52 (2012) 112–124

Fig. 3. Alternative 2, reverse pairing for region I, and also close to the optimal structure for region II.

(4) is negative, whereas we found from the dynamic model in (2) for the initial increase in the CO2 recovery is an initial decrease
that the diagonal elements are dominant at higher frequencies cor- in the CO2 vapor mole fraction in the top of the absorber, which is
respond to having a positive gain at steady state. The sign change expected. However, on a longer time scale, with an increased feder-
can be explained by Fig. 5, where we see that y1 (CO2 recovery) ini- ate to the stripper column (u1 ) and a constant reboiler duty (u2 ), the
tially increases in response to a step increase in u1 (recycle amine CO2 concentration in the bottom of the stripper increases and starts
flowrate). This agrees with the dynamic model Gdyn. . The reason “filling up” the absorption column with CO2 . This takes a long time
because of the large holdup in the absorber (see Fig. 6). Finally, the
CO2 “breaks through” at the top of the absorber column and after
almost 170 min the CO2 recovery (y1 ) starts decreasing and keeps
decreasing until about 250 min. There is a second smaller decrease
in y1 about 70 min later. To actually get a negative gain for y1 , we
need the change in recycle amine flowrate (u1 ) to be sufficiently
large, so this is partly a nonlinear effect. The dynamic model Gdyn.
is based on simulations on an intermediate time scale and this is
why the reversal of the sign of the gain was not identified.

Fig. 5. Response in CO2 recovery (y1 ) to step change in recycle amine flowrate (u1
Fig. 4. RGA number of two different alternatives in pairing. by 5%).
 M. Panahi, S. Skogestad / Chemical Engineering and Processing 52 (2012) 112–124 117

Fig. 6. Sluggish dynamic response of the CO2 mole fraction due to large holdup on absorber trays.

The steady-state RGA suggests “reverse pairing” which is pairing (“Alternative 1”) should be used in region I because the CVs
referred to as Alternative 2. and the MVs are close, but we need to reconfigure the loops if the
Based on pairing rule 2 and the model Gss , we would conclude reboiler duty reaches its maximum (region II). Alternative 2, on the
that the diagonal pairing (Alternative 1) should not be used. How- other hand would work in both regions without any reconfigura-
ever, recall that the reason for RGA rule 2 is to avoid instability tion. However compared to Alternative 1 the dynamic performance
if one of the loops is no longer working, which in this case will in region I will be poorer as the associated MVs are not close-by.
occur when the reboiler duty saturates. In summary, the diagonal This is confirmed by dynamic simulations later.

Fig. 7. Alternative 3, proposed decentralized control structure for region II [1].

118 M. Panahi, S. Skogestad / Chemical Engineering and Processing 52 (2012) 112–124

Fig. 8. Alternative 4, proposed structure for combined regions I and II (modified of Alternative 2).

3.2. Operation in region II (Alternative 3) In Alternative 2, the liquid inflow to the stripper, which actu-
ally has a much more direct effect on the stripper temperature,
Let us next consider operation in region II. The “optimal” solu- is manipulated to control the level of the absorber. A modification
tion for region II, is to let the reboiler duty stay at its maximum, (Alternative 4) is to change the pairings for these two loops, which
“give up” controlling the CO2 recovery, and move the stripper tem- can be viewed as moving the location of the given flow in the recycle
perature from tray 16 to tray 13 [1] (Alternative 3, Fig. 7). Similar loop from the inlet to the outlet of the absorber (Fig. 8).
to Alternative 2, the stripper temperature is controlled using the
recycle flow which is the only available “free” MV in region II.
For strictly optimal operation in both regions I and II, one could
4. Performance of alternative control structures
use same supervisory logic that switches between Alternatives 1
and 3. It would switch from Alternative 1 to Alternative 3 when the
In this section we analyse the four alternatives, as well as mul-
reboiler duty saturates and it would switch back from Alternative
tivariable control (MPC), using dynamic simulation. We observed
3 to Alternative 1 when CO2 recovery passes 95.26%.
some discrepancy in the nominal steady-state in dynamic mode
However, note that Alternative 3 (region II) is actually very close
versus the steady-state mode. For example, for the same nom-
to the “reverse pairing” (Alternative 2) for region I. This suggests
inal steady state CV values, the reboiler duty in the dynamic
that Alternative 2 could also be used to handle region II, although
mode is 1074 kW while in the steady state mode the reboiler duty
it would involve a small loss because the stripper temperature is
is 1161 kW. This is partly due to the dynamic simulation being
not at its optimal location in region II. With Alternative 2, switching
pressure driven so that the column pressure profiles are slightly
between regions I and II would simply be to give up control of CO2
different. In addition, the UniSim dynamics solver may be using
recovery when the reboiler duty saturates.
approximate thermodynamic models. Fortunately, from a con-
trol structure synthesis perspective, the difference in the nominal
3.3. Alternative 4 (regions I and II) steady states is not very important as the first constraint to become
active (reboiler duty) as throughput increases, remains unchanged
So far Alternative 2 is the best structure for the combined regions in both modes.
I and II, and we want to improve on it. In Alternative 2, the recycle As in part I, there are maximum capacities compared to nominal
amine flowrate (V-8) is manipulated to control the stripper tem- values for the reboiler duty (+20%), cooler (+50%) and pumps (+40%).
perature. However, this control loop has a large effective delay These constraints are even more important in the present dynamic
so performance is relatively poor (see dynamic simulations later). study.
 M. Panahi, S. Skogestad / Chemical Engineering and Processing 52 (2012) 112–124 119

Table 3
Tuning parameters (Alternative1).

Control loop Closed loop time constant,  c (min) Kc  I (min)

CO2 recovery (y1 ) with recycle amine flowrate (u1 ) 0.38 0.315 3.04
Temp. 16 in the stripper (y2 ) with reboiler duty (u2 ) 0.30 10.53 2.4

4.1. Alternative 1 (region I) no. 16 (Fig. 9c) in the stripper is not controlled in its setpoint, but
with a further increase in the flue gas flowrate (at t = 358 min), the
We first consider the diagonal pairing for region I (Fig. 1) and stripper temperature drops further (t = 361 min) resulting in insuf-
look at the performance (dynamic behavior) when there are large ficient CO2 stripping which builds up in the amine recirculation
disturbances. All the controllers were tuned using SIMC method loop. This necessitates the more amine flow in the absorber to meet
[12] with tuning parameters in Table 3. The flowrate of flue gas is the desired CO2 recovery which saturates the pump 1 (Fig. 9d) at
increased gradually in steps of 5% from 0% to +25% compared to the t = 437 min. At t = 600 min the recycle valve V-8 (u1 ) becomes fully
nominal flowrate (Fig. 9a). From Fig. 9 we see that the control struc- open (Fig. 9e) and we get an unstable system where the CO2 recov-
ture behaves very well in region I (up to +20%) with tight control of ery can no longer be maintained at 95.26% (Fig. 9f) resulting in large
the CO2 recovery and stripper temperature. When the flowrate of objective function value (Fig. 9g). This is as expected from the ear-
flue gas increases +20% (at t = 269 min), the reboiler duty saturates lier RGA analysis. In summary, pairing Alternative 1 can only handle
at t = 273 min (Fig. 9b), signifying a transition to region II. Initially, increased flue gas flowrates of up to +20% (region I), at which point
it seems that the system is stable although the temperature of tray the reboiler duty (u2 ) saturates.

Fig. 9. Dynamic simulation of pairing Alternative 1 (Fig. 1). The structure handles increase in flue gas flowrates of up to +20% (region I).
120 M. Panahi, S. Skogestad / Chemical Engineering and Processing 52 (2012) 112–124

Fig. 10. Region II: Alternative 3 (Fig. 7) handles increase in flue gas flowrate to +43%, but it does not cover region I.

4.2. Alternative 3 (region II) 4.3. Alternative 2 (regions I and II)

We here consider the “optimal” structure for region II. (“Alter- The simulations in Fig. 12 show that this alternative can handle
native 3”, Fig. 7). In the simulation, for a closed loop  c = 2.33 min, increase of the flue gas starting from the nominal value (0%) and
SIMC method gives Kc = 0.314,  I = 14 min as PI tuning parameters. up to +42% (Fig. 12a). The reboiler duty (u2 ) saturates (Fig. 12b)
Fig. 10 shows how this structure handles feed flowrate changes when the flue gas flowrate increase is +20%, at which point the CO2
with operation within region II, starting from 20% above the nom- recovery control is given up, and at +42% the minimum CO2 recov-
inal (where the process enters region II) with steps of +5%. In ery of 80% (Fig. 12f) is met. Note that we in both regions control y2
this case, we do not attempt to control CO2 recovery and it drops (temperature of tray no. 16 in the stripper). This is the best choice
gradually from 95.26% to 80%, which is the minimum allowable in region I, and in region II it is close to the best self-optimizing
CO2 recovery. This happens when the feed flowrate increase is variable which is y3 (temperature of tray no. 13 in the stripper).
+43% (Fig. 10a). As shown in Fig. 10c, the stripper temperature Thus, it is not surprising that if Alternative 2 can handle changes
is well controlled and hardly affected by the feed flow distur- of +42% which is close to the region II optimal at +43% (using
bance. Alternative 3).
In general, any active constraint should be considered as a dis- The advantage of Alternative 2 compared to optimal structure
turbance [2], thus we changed the fixed (saturated) reboiler duty (which is to use Alternative 1 in region I and Alternative 3 in
by ±10% and the dynamic responses in Fig. 11 shows that also this region II) is that we do not need switching of CVs. The main dis-
disturbance is handled well with stripper temperature deviation of advantage is that we have more interactions than Alternative 1
less than 3 ◦ C. in region I. This is seen by comparing the simulations in Fig. 9
 M. Panahi, S. Skogestad / Chemical Engineering and Processing 52 (2012) 112–124 121

Fig. 11. Region II: the structure in Fig. 7 also handles disturbance in reboiler duty which is the active constraint.

(Alternative 1) and 12 (Alternative 2) in region I. The CO2 recov- with 1 ◦ C in Fig. 12c). In region II, using Alternative 2, there is a
ery is not as tightly controlled (compare the max. deviation of 0.4% minor loss compared to Alternative 3 (compare Figs. 10g and 12g)
in Fig. 9f with 3% in Fig. 12f) and also the stripper temperature and also here the temperature control is comparable (compare
shows larger variations (compare max. deviation of 0.3 ◦ C in Fig. 9c Figs. 10c and 12c).

Fig. 12. Dynamic simulation of alternative 2 (reverse pairings) in combined regions I and II. The structure handles the increase in flowrates of flue gas to +42%.
122 M. Panahi, S. Skogestad / Chemical Engineering and Processing 52 (2012) 112–124

Fig. 13. Dynamic simulation of pairings alternative 4 (modified reverse pairings) in combined regions I and II.

4.4. Alternative 4 (regions I and II) models and the final responses with MPC are shown in Fig. 14. For
feed values greater than +20%, when the reboiler duty saturates
Alternative 4 (Fig. 8) involves repairing the absorber level con- (region II), instead of controlling the CO2 recovery at its setpoint
trol. As shown in Fig. 13, this structure handles increase in flue gas value (95.26%), we put less emphasis on controlling the CO2 recov-
flowrate up to +42% like in Alternative 2 but with tighter control of ery by introducing a range with a lower bound of 80% for this CV. The
stripper temperature. By comparing Figs. 12c and 13c, we see that result is that RMPCT controls the temperature of tray no. 16 at its
the stripper temperature deviation decreases from 1 ◦ C to 0.2 ◦ C setpoint and gives up controlling the CO2 recovery. When the flue
while control of CO2 recovery remains the same. Therefore, Alter- gas flowrate reaches +42% the minimum allowable CO2 recovery
native 4 may be the best structure for practical implementation. of 80% is met and there are no degrees of freedom left and further
increase of flue gas flowrate is infeasible.
4.5. Multivariable controller (regions I and II) We finally compare the economical performance of all the
structures. From the objective function in Figs. 10g (Alterna-
Finally, we consider a multivariable controller (MPC) obtained tive 3), 12g (Alternative 2), 13g (Alternative 4) and 14g (MPC) we
using the RMPCT (Robust model predictive controller) from Hon- find that there is only a small difference. At the final steady state, the
eywell [3]. The aim is to compare its performance with the previous objective function values are 3.38, 3.39, 3.39 and 3.39 (USD/ton of
decentralized PI controllers. This MPC includes 2 CVs (y1 and y2 ), 2 flue gas) respectively. This implies that we can use both Alternative
MVs (recycle lean amine flowrate, u1 and reboiler duty, u2 ) and 2 2, Alternative 4 or MPC to control the system even in the presence of
disturbances (flowrate of flue gas and its CO2 composition). We large flowrates of flue gas. Note that the design and implementation
used Honeywell’s Profit Design Studio to identify the dynamic of reverse pairings with decentralized controllers are simpler and
 M. Panahi, S. Skogestad / Chemical Engineering and Processing 52 (2012) 112–124 123

Fig. 14. Performance of RMPCT tuned in region I, when extended to region II, the structure handles increase in flue gas flowrates of +42%.

cheaper than MPC which needs model identification etc. although performance comparable to MPC, and is proposed as the best alter-
responses time are comparable. native because of its much simpler implementation.
Of all the alternatives considered, Alternative 4 is proposed as
the best structure for the CO2 capturing process, studied here. References

[1] M. Panahi, S. Skogestad, Economically efficient operation of CO2 capturing

process. Part I. Self-optimizing procedure for selecting the best controlled
5. Conclusions variables, Chemical Engineering and Processing: Process Intensification 50 (3)
(2011) 247–253.
[2] S. Skogestad, Control structure design for complete chemical plants, Computers
Using the bottom-up part of the plantwide control procedure, and Chemical Engineering 28 (2004) 219–234.
we obtained alternative control structures that implement the opti- [3] UniSim Design R380, Honeywell Company, 2008.
mal controlled variables (CO2 recovery in the absorber and the [4] Y.-J. Lin, T.-H. Pan, D.S.-H. Wong, S.-S. Jang, Y.-W. Chi, C.-H. Yeh, Plantwide
control of CO2 capture by absorption and stripping using monoethanolamine
temperature of tray 16 in the stripper) from part I. Alternative
solution, Industrial Engineering and Chemical Research 50 (3) (2011)
1 (diagonal pairing of the best CVs found in region I with close- 1338–1345.
by MVs: recycle amine flowrate and reboiler duty) can handle [5] S. Ziaii, G.T. Rochelle, T.F. Edgar, Dynamic modeling to minimize energy use
for CO2 capture in power plants by aqueous monoethanolamine, Industrial
flue gas flowrates of up to +20% (region I). The reverse pairings
Engineering and Chemical Research 48 (13) (2009) 6105–6111.
(Alternative 2) or MPC handle flow values up to +42%. Due to the [6] H.M. Kvamsdal, J.P. Jakobsen, K.A. Hoff, Dynamic modeling and simulation of a
large delay between the paired MVs and CVs in Alternative 2, a CO2 absorber column for post-combustion CO2 capture, Chemical Engineering
modified structure (Alternative 4) is proposed where the recycle and Processing: Process Intensification 48 (1) (2009) 135–144.
[7] A. Lawal, M. Wang, P. Stephenson, H. Yeung, Dynamic modelling of CO2 absorp-
amine flow manipulator is moved from the inflow to the outflow tion for post combustion capture in coal-fired power plants, Fuel 88 (12) (2009)
of the absorber. This simple structure (Alternative 4) has dynamic 2455–2462.
124 M. Panahi, S. Skogestad / Chemical Engineering and Processing 52 (2012) 112–124

[8] M. Panahi, M. Karimi, S. Skogestad, M. Hillestad, H.F. Svendsen, Self- [10] S. Skogestad, I. Postlethwaite, Multivariable Feedback Control Analysis and
optimizing and control structure design for a Co2 capturing plant, in: Design, 2nd ed., 2005.
Proceedings of the 2nd Annual Gas Processing Symposium, 2010, pp. 331–338, [11] Honeywell Profit Design Studio R310, Honeywell Company, 2007.
doi:10.1016/S1876-0147(10)02035-5. [12] S. Skogestad, Simple analytic rules for model reduction and PID controller tun-
[9] E.M.B. Aske, S. Skogestad, Consistent inventory control, Industrial Engineering ing, Journal of Process Control 13 (2003) 291–309.
and Chemical Research 48 (24) (2009) 10892–10902.